There are Major Flaws in 538's Do-It-Yourself Election Model
The video uses Florida in a very flawed manner.
Do It Yourself
Let's change one state, Florida.
- A change in Florida dramatically changes the forecast elsewhere, flipping Arizona and North Carolina as well.
- Arizona went from 67-33 Biden to 74-24 Trump.
- North Carolina went from 67-33 Biden to 69-31 Trump.
The inherent problem is 538 made no distinction between Trump winning by 0.5% or Trump winning by 2.5%.
To the interactive model, a win is a win.
Moreover, the model appears to assume that we know the result in Florida, but have nothing but 538's projections for the other states.
Arizona may very well be decided on election night. We will know something about North Carolina as well.
Changing Florida and Arizona
In the above example, I set Florida and Arizona, one for Trump and one for Biden. That's not an unlikely possibility, in any direction.
Now the results are back to 83-16 in favor of Biden.
Early Vote Counting
The video example makes it appear as if this all comes down to Florida again. It doesn't.
So no, we cannot say that if Trump wins Florida, he has a 40% chance of winning.
- The margin of Trump's alleged Florida win certainly matters. 538's interactive model says a win is a win.
- We will not magically see Florida in a vacuum.
- The video example over-emphasizes Florida. It is not that simple.
In practice, we will have partial results elsewhere and perhaps even full results in Arizona.
The interactive model seriously needs work. I suggest it be reworked to put in custom percentages instead of projecting winners.